PyelogP：基于能量的黏土沉积物先期固结压力自动化确定方法
PyelogP: Automated Energy-Based Determination of Preconsolidation Pressure in Clay Deposits
浏览论文内容
中文总结 AI 辅助
本文提出开源Python库PyelogP，利用应变能法自动确定黏土沉积物的先期固结压力，结合样条插值、拐点检测和线性回归，在22个数据集上验证，结果与已发表值高度一致且计算高效。
中文摘要 AI 辅助
在一维固结（固结仪）试验中估算先期固结压力（$\sigma'_p$）对岩土工程中的沉降分析至关重要。传统的图形方法，如Casagrande方法，在解释软黏土和粉土沉积物扰动试样典型的圆滑$e$-log($P$)曲线时可能引入不确定性。本文介绍了PyelogP，一个开源的Python库，旨在使用Becker等人（1987年）提出的应变能方法自动且可重复地计算$\sigma'_p$。该算法结合了自然三次样条插值、通过Kneedle算法进行拐点检测，以及在功-压力空间中的分割点线性回归。物理信息阈值，包括超固结比限制和二阶导数最大值（${d^2 e}/{d(\log \sigma')^2}$），被纳入以建立屈服前和屈服后的拟合边界。PyelogP的性能通过一组22个实验固结数据集进行评估，这些数据集涵盖了各种黏土沉积物，包括Saint-Alban黏土和旧金山老湾黏土。结果表明，与已发表值高度一致（$R^2$ = 0.912，RMSE = 0.374，MBE = -0.080），而$O(N^2)$算法对于典型的固结仪数据集每条曲线仅需几毫秒，对于最大的数据集也只需不到150毫秒。
英文摘要
Estimating the preconsolidation pressure ($σ'_p$) from one-dimensional consolidation (oedometer) tests is critical in geotechnical engineering for settlement analysis. Traditional graphical methods, such as the Casagrande procedure, may introduce uncertainties, particularly when interpreting rounded $e$-log($P$) curves typical of disturbed specimens of soft clays and silt deposits. This paper introduces PyelogP, an open-source Python library designed to calculate $σ'_p$ using the strain-energy method proposed by Becker et al. (1987) as an automated and reproducible alternative. The algorithm combines natural cubic spline interpolation, knee-point detection via the Kneedle algorithm, and split-point linear regression within the work-pressure space. Physically informed thresholds, including overconsolidation ratio limits and second-derivative maxima (${d^2 e}/{d(\log σ')^2}$), are incorporated to establish pre-yield and post-yield fitting boundaries. The performance of PyelogP is evaluated against a suite of 22 experimental consolidation datasets covering various clay deposits, including Saint-Alban clay and San Francisco Old Bay Clay. The results demonstrate strong agreement with the published values ($R^2$ = 0.912, RMSE = 0.374, MBE = -0.080), while the $O(N^2)$ algorithm requires only a few milliseconds per curve for typical oedometer datasets and less than 150 milliseconds for the largest datasets.
